How Sentry divided its MTTR
by 5 with AI triage

Sentry runs its own error-monitoring platform internally. With more than 10,000 alerts per week generated by their infrastructure, the SRE team spent most of its time triaging notifications, 78% of which turned out to be noise (false positives, already-resolved alerts, events with no user impact).

This alert fatigue led to longer response times on genuinely critical incidents, and to gradual demotivation across the engineering team. MTTR (Mean Time To Resolution) had doubled in 18 months despite tooling investments.

In 9 days, the SRE team deployed an EzyAgents agent connected to their monitoring stack (PagerDuty, Datadog), their codebase (GitHub) and their incident history. The agent analyzes every incoming alert in real time and makes a routing decision.

• Automatic classification of every alert: critical, major, minor, noise
• Correlation with the history of similar incidents for immediate context
• Notifies only the relevant on-call engineer, with full context
• Runbook suggestions and corrective actions based on past resolutions
• HITL approval before any automatic rollback in production
• Automatic post-mortem generated in Notion after every P1/P2 incident is resolved

From the first week, the volume of notifications sent to engineers dropped by 80%. The agent correctly classifies 94% of alerts without human intervention. MTTR went from 47 minutes to 9 minutes on average for P1 incidents.

The SRE team reclaimed an average of 12 hours per engineer per week, reassigned to proactive reliability projects. Recurring incidents dropped by 60% in three months thanks to automated post-mortems and systematic root-cause identification.